Dual-Stage Stacking Machine Learning Method Considering Virtual Sample Generation for the Prediction of ZIF-8' BET...

Area of Science:

• Materials Science
• Computational Chemistry
• Environmental Engineering

Background:

• Metal-organic frameworks (MOFs) are increasingly used in gas and wastewater treatment.
• Accurate and rapid assessment of BET specific surface area is crucial for MOF applications.
• Current experimental methods are often time-consuming and costly.

Purpose of the Study:

• To develop a novel computational model for predicting the BET specific surface area of MOFs.
• To improve the efficiency and accuracy of MOF characterization.
• To reduce the reliance on expensive experimental techniques.

Main Methods:

• A dual-stage stacking model was proposed, incorporating Gaussian mixture model-virtual sample generation (GMM-VSG) technology.
• Machine learning models including Bayesian regression, AdaBoost, random forest, and XGBoost were evaluated.
• A two-stage stacking model was constructed using the top-performing models and linear regression.

Main Results:

• The dual-stage stacking model achieved an R-squared value of 0.974 on virtual and real samples.
• Feature importance analysis guided adjustments to experimental conditions.
• The final prediction accuracy for BET specific surface area reached 0.943.

Conclusions:

• The proposed GMM-VSG-enhanced dual-stage stacking model offers a highly accurate and efficient method for BET specific surface area prediction.
• This approach can significantly accelerate the development and application of MOFs in environmental remediation.
• The study provides a valuable computational tool for materials scientists and engineers.